How can I remove shake from images like this one?

Question

Sometimes I have images where the shake motion is not linear but it can be guessed from lights and similar point-like details. I am aware that this photo is low quality, it's simply the latest one I wanted to improve.

How can I remove that shake? I'm working on Windows.

I already found this promising MATLAB script but I don't have Matlab.

Update

Using blurity with a custom blur model (a single light trail) isolated manually and loaded separately I obtained the following result.

If there were fewer JPG artifacts, the result would have been quite good!

Answer 1

Try these links:

• https://www.blurity.com/
• http://www.crunchytricks.com/2016/05/tools-to-unblur-photos.html

However, the best way to avoid blurred pictures is to practice better photography and avoid hand shakes while clicking. Try to take pictures in better light if you're just using a mobile camera - OR use a better camera like a DSLR - but that takes you on a different path of learning photography altogether.

Answer 2

Photoshop CC has a Shake Reduction filter to try to remove motion blur. However, your particular example image might not be well suited to the blur removal feature. From Adobe's help page about the Shake Reduction filter:

The camera shake reduction feature works best with decently lit still camera images having low noise. The following types of still images are particularly suitable for shake reduction:

• Indoor or outdoor images captured using a lens with a long focal length
• Indoor images of a static scene taken with a slow shutter speed and no flash

However, it's an impressive feature, and for the images best suited for it, the filter does an impressive job.

Answer 3

I simply don't try to do anything with it. I have actually captured many of these type of images, of course, unitentionally, and posted them with the caption Shakiness of the Modern Life! and believe me, those pics got more than 100 likes!

Answer 4

MIT has done some work on this, you have already found it: https://cs.nyu.edu/~fergus/research/deblur.html , and indeed it points to Matlab code. There exist open source replacements for Matlab, but not if you also need expensive image processing libraries.

Beware, de-blurring is a de-convolution, essentially a high-pass filter. Any noise and compression artefacts in the image will be amplified without mercy.

An interesting development is the "flutter shutter" or "coded aperture", where the shutter opens multiple times so that the blur is easier to invert (no deep zeroes in the transfer function). This is an application of "computational photography". You are more likely to find this in an expensive smartphone than in a plain old camera.